This is a python implementation of algorithms and experiments presented in the paper "Graphical modelling in continuous-time: consistency guarantees and algorithms using Neural ODEs".
Graphical modelling is the problem of defining and piecing together associations in data to infer the underlying structure among a system of variables. This project considers score-based graph learning for the study of dynamical systems. The proposal is a score-based learning algorithm based on penalized Neural Ordinary Differential Equations that we show to be applicable to the general setting of irregularly-sampled multivariate time series.
This project uses pytorch and torchdiffeq. For full list of dependencies see requirements.txt or environment.yml (for conda). In order to run the model and the paper experiments, install the dependencies from the appropriate file.
To get started, check Tutorial.ipynb which will guide you through graphical modelling in continuous-time from the beginning.
For the experiments, see Paper_experiments_Lorenz.ipynb and Paper_experiments_Rossler.ipynb.